
No code changed in two months, and it still gains 10 stars a day: 3dviz-pro-max, whose traffic is it borrowing?
What it is A manual that teaches coding agents to make 3D scenes look good (223 recipes, 37 examples), plus a script that lets the agent screenshot its own output and check it.
Why it took off The author is the top contributor to ui-ux-pro-max-skill (147k stars). The new project reused the "Pro Max" name. On launch day, a big account's README put its banner at the top, and it never came down. On top of that, the author posted it on Facebook, and Chinese GitHub recommendation accounts reshared it.
BHow you could follow this Don't build a sprawling 3D skill like it did. Pick a narrow topic that's instantly understandable and shareable from a screenshot (like cutaway machines or interactive models), keep the screenshot self-check, write the README like an asset card, and on launch day, go after Chinese and English GitHub recommendation accounts directly.
- Data (updated daily) · Repo created 2026-09-10 · 1,433 stars · 7 days +177
GitHub · stars Data 2026-10-09 → 2026-11-20 (43 days, 41 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
A 3D visualization skill for Claude Code and Codex that picked up 1,146 stars in 60 days, yet its code stopped changing the day after launch. The stars weren't earned by its features. They mostly came from borrowing traffic from a 147k-star repo. Below: what it borrowed, what you can't borrow, and what the 3D skills that grew much faster at the same time did right.
The Sep 10 Facebook post
2026 Sep 10: Viet Tran (on GitHub as viettranx), founder of the Ho Chi Minh City programming bootcamp brand 200Lab, posted on his Facebook page. The page has 21k followers. The post opens with "Share skill mới: 3D Visualize Pro Max," followed by "Xưa có UI UX Pro Max, nay có 3D Visualize Pro Max cùng mindset" ("Once there was UI UX Pro Max, now there's 3D Visualize Pro Max with the same mindset"). The link was in the comments. The post was shared 253 times, and a Vietnamese developer opened the first PR the same day.
That line, "Once there was UI UX Pro Max," is the key to the whole thing. ui-ux-pro-max-skill is one of the best-known repos in the agent skill scene: 147k stars, still gaining about 3,000 a week today. Viet Tran is its biggest contributor. The same afternoon (13:34 Vietnam time), another core maintainer of that repo, Duy Nguyen (mrgoonie), pushed 4 changes adding a "Check Out Our New Skill: 3Dviz Pro Max" block to the very top of the README, with a logo, an animated GIF, and a website link. Two months later the banner is still there.
What it is
In one line: a manual for coding agents on how to make 3D look good, plus a screenshot script.
You tell the agent, "Use 3dviz-pro-max to build me a fantasy village you can walk through, pick a distinctive art style, add a few animated critters, then check the result." It follows a ten-step process: set the art direction, work out the object structure, pull ready-made recipes and components from the library, build a first version, run it and screenshot it, revise against the screenshot, and finally write up what it did and what actually shows on screen.
The library numbers sit at the very top of the README: 223 recipes, 440 knowledge entries, 22 component sets, and 37 runnable examples, covering 24 domains, from fantasy villages and product teardowns to heart cross-sections, linear algebra, and robot kinematics. There is also a standalone site, 3dviz.dev. MIT license, free.
The screenshot step is what actually hooks people. In Sep 14, GitHubDaily on Chinese X introduced it by opening with: "Have an agent build a 3D scene with Three.js, and you often get a few cubes and a light, or a black screen, while it insists it's done." Then: "What sets it apart is that it forces the agent to look at the image it rendered." Nearly every repost after that repeated this line.
How the stars came
| Timeline | What happened | Stars gained that day |
|---|---|---|
| 09-10 | Author's Facebook first; a big-account README banner; the team's site launched a product page and the official X account posted | Day-one ~130 (inferred: third-party sites weren't tracking yet, so the early star count is back-calculated from the 09-25 star count at 545) |
| 09-11 | Chinese recommendation account @QingQ77 posted, 1,935 views; last commit to the code | +40 |
| 09-14 | GitHubDaily posted, 10.3k views, 122 likes, 160 bookmarks | +45 |
| 09-15 | Aftershocks; that day it hit #12 on Trendshift's JavaScript chart | +139, the highest point of the whole curve |
| 09-15 to 09-25 | English newsletters, Substack, a LinkedIn summary, and a French X post picked it up; YouTube mini-accounts got a few dozen to a few hundred views each | +10 to +45 per day |
| Late September to 11-06 | No new code, no new posts | +4 to +30 a day, mostly around 10 |
| 11-07 to 11-08 | The TikTok repost account Github Signals (26k followers) posted a video: 3,808 views, 169 bookmarks; a compilation video titled "12 free Claude skills" got 963 views; plus a Chinese X post | +24, +66 |
Daily numbers come from Trendshift and GitHub daily snapshots, as of 11-09.
It never made Hacker News or Product Hunt, and I found no Reddit threads or paid promotion. The entire reach was Vietnamese developers on Facebook, a Chinese X GitHub recommendation account, a string of English newsletters and directory sites, and then, two months later, a TikTok with a few thousand views. Against that backdrop, 1,100 stars is a lot.
Why it keeps growing without code changes
The last commit was Sep 11. In the nearly two months since, there have been no updates and the author hasn't posted again, yet the stars keep steadily climbing 8 to 15 a day.
The most plausible explanation is that banner (inferred: only the maintainer can see where repo traffic comes from, so outsiders can't verify it). A repo with 147k stars still gains about 3,000 stars a week, and every passerby adds to that. 3dviz gains roughly 10 stars a day, about 2.5% of what the big account gains per week. In other words, the banner's conversion rate isn't high, but it never stops. That's the foundation of 3dviz: an always-on ad slot, free, owned by the team.
Another number to compare: skills.sh has recorded only 176 installs, and the ratio of stars to installs is roughly 6.5 to 1 (manual copies and plugin-marketplace installs aren't counted). Many people click the star; far fewer actually install and use it.
Meanwhile, others are growing much faster
"Let a coding agent do 3D" has been a hot niche this year, with several repos passing ten thousand stars between July and October. 3dviz is a mid-sized player in there.
| Repo | Launched | Stars (11-09) | What it does |
|---|---|---|---|
| wheresryan22/anatomy | 10-08 | 26k, +7 over the last 2,268 days | Does one thing only: isometric cutaway illustrations of real machines and complex systems |
| img2threejs | 07-15 | 21k | Takes a reference image and rebuilds it as a Three.js model |
| oso95/scroll-world | 07-06 | 11k | Turns a brand into a scrollable 3D landing page |
| CloudAI-X/threejs-skills | 01-19 | 4,076 | A Three.js API cheat sheet, no opinions on aesthetics |
| elvarofficial/teardown | 10-30 | 2,657, +7 over the last 1,784 days | Turns any machine into a 3D simulation you can take apart |
| achimala/dream-loop | 09-07 | 1,853 | Blender plus image generation plus subagent review; launched 3 days before 3dviz, also emphasizing "look at the picture" |
| Moh4696/open-3d-skills | 10-31 | 307, +7 over the last 201 days | 8 3D web effect skills, also with screen-recording checks |
| viettranx/3dviz-pro-max | 09-10 | 1,146 | 24 opportunities covered, with recipes and screenshot acceptance |
The contrast is stark. teardown passed 3dviz's two-month total in 10 days; anatomy hit 26k stars in a month, more than twenty times its total. The ones that surge are narrow in topic and instantly understandable from one image: a machine cut open, a machine you can take apart piece by piece. 3dviz's pitch is "it can do anything." The animated fantasy village it puts at the top of the README carries an author note saying "recorded from an old project before the skill existed, not output from this skill." The first screen, where it should hook people most, is instead carrying a disclaimer.
"Make the agent look at its own output" isn't unique to it either: dream-loop launched 3 days earlier, and open-3d-skills also does screen-recording checks. The line gives reposters something easy to talk about, but it can't carry star counts at this scale. Star counts at this scale are decided by distribution.
What you can learn, and what you can't
Can't copy: the permanent banner atop a 147k-star repo, the author's Facebook with 21k followers, and the "Pro Max" series name. These three things were the source of 3dviz's roughly 130 stars on day one and 10 stars a day after that. Without a comparable big account, you can't reproduce its curve.
Can copy:
- Write the README as an asset card. A table of numbers ("how many X we have") plus one pain point ("the agent says it's built, but it's all black"). GitHubDaily, the TikTok repost account, and the English newsletters copied both nearly verbatim. Recommendation-account editors are busy; if you write the copy well, they'll use it as is.
- Launch in a sequence. The author's personal accounts, the team account, and linked repos went live the same day; three or four days later, the Chinese recommendation accounts picked it up, and two months on, repost accounts added another wave. Chinese GitHub recommendation accounts really move developer tools: one post with 10k views produced the highest day on the whole curve.
- A demo gallery you can open directly. 37 working examples plus a standalone site give directory sites and repost accounts something to screenshot.
If it were you
Verdict: B. Copying 3dviz to build a "draw anything" 3D skill isn't worth it; its stars are mostly borrowed, and a broad topic on its own doesn't drive stars. What's worth doing is a narrow topic within the same niche: pick one 3D treatment that's instantly understandable and shareable as a screenshot, and build it as a skill. anatomy and teardown have already shown this approach can still draw crowds in October and November.
Who it suits: developers who know Three.js or Blender and have their own sense of aesthetics; better if you know a specific vertical (medicine, machinery, architecture, chemistry, geography), so you can pick the thing people in that field want to see and agents consistently draw badly.
Who it doesn't suit: people who want a big, all-encompassing knowledge base; people with no launch channels who aren't planning to actively pursue recommendation accounts; people who can't show real output from their own product on the first screen.
How to build it: one skill does one thing, with that thing named right in the title ("turn any machine into a 3D you can take apart" is far more memorable than "3D Visualization Pro"). Keep screenshot verification; it's a real user pain point and a ready-made talking point. Put a GIF the skill itself produced on the first screen of the README, with a number table and a pain-point line beneath; support Claude Code, Codex, and Cursor at the same time.
How to get stars: On launch day, post from your own X, Facebook, or Jike account; prepare ready-to-use copy in both Chinese and English and send it directly to Chinese recommendation accounts like GitHubDaily and QingQ77 and to English GitHub newsletters; submit to skill directories like skills.sh; film a 30-second vertical video going from "one sentence" to "the picture appears," for GitHub repost accounts on TikTok and Shorts. If you or a friend has a repo with a thousand-plus stars, put a banner at the top of its README; this is the most durable move 3dviz has proven.
When to get in early: this niche saw three repos pass ten thousand stars between July and October (img2threejs, scroll-world, anatomy), and good topics are getting claimed one by one. Try to ship before the end of 2026. If someone has already reached several thousand stars on the same topic, switch to a different one.
Signs you've misread it: if new narrow 3D skills launched over the next month or two stay under a few hundred stars (open-3d-skills is a reference point, +7 over the last 201 days), the wave has receded, or agents' built-in 3D abilities are already good enough. If Anthropic or OpenAI build screenshot self-checks and 3D templates into official skills or coding agents, this category's value will be directly absorbed.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one wheresryan22/anatomy Open source 2026-10-08 25,985★, +2,268 over 7 days Focused on isometric cutaway illustration: a narrow topic with strong visual impact. 3dviz is general 3D scenes: a broad topic, but without a first screen that stops you cold. img2threejs/img2threejs Open source 2026-07-15 20,924★ Rebuilds procedural Three.js models from reference images; the input is an image. 3dviz's input is a sentence, and it covers entire scenes. ahujasid/mcp-for-blender Open source 2025-03-07 33,652★ An MCP tool interface for driving Blender; no aesthetic knowledge base, and no screenshot verification. oso95/scroll-world Open source 2026-07-06 10,883★ Focused on scrollable 3D brand landing pages, positioned as a marketing-page tool. CloudAI-X/threejs-skills Open source 2026-01-19 4,076★ A Three.js API cheat sheet: covers the API, not aesthetics, and has no screenshot checks. elvarofficial/teardown Open source 2026-10-30 2,640★, +1,784 over 7 days A single concrete approach (simulations of machines you can take apart). It passed 3dviz's two-month total in 10 days. achimala/dream-loop Open source 2026-09-07 1,846★ Launched 3 days earlier; uses Blender, image generation, and subagent review to produce the output, also emphasizing "look at the picture." Moh4696/open-3d-skills Open source 2026-10-31 306★, +201 over 7 days Focused on 3D web effects, also with screen-recording checks. Anthropic's official skills / Claude Code frontend capabilities Big-company features From ¥2025 anthropics/skills, about 188k★ General frontend design, with no 3D-specific recipes or knowledge base. - Where users come from
- Verified README banner on a large repo: 09-10 mrgoonie added a "Check Out Our New Skill: 3Dviz Pro Max" block to the README of nextlevelbuilder/ui-ux-pro-max-skill, which has 147k stars (commits 53b8e69 and 4 more), and it was still there as of 11-09. There is also an NLB website product page and an @nlb_io X post.
- Verified the author's Facebook launch: Viet Tran's profile with 21k followers posted "Share skill mới: 3D Visualize Pro Max" on 09-10; the original text says "previously there was UI UX Pro Max," with 253 shares and 24 comments.
- Verified Chinese X GitHub recommendation accounts: GitHubDaily posted on 09-14 10:00 UTC (10.3k views, 122 likes, 160 bookmarks), immediately followed by 09-15 +139; QingQ77 09-11 (1,935 views); Jolyne_AI, hank_aibtc, Smartpigai, denziideng (11-07)
- Verified TikTok/Shorts/Reels repost accounts: Github Signals (26k followers) posted a video at 11-07 20:14 UTC, 3,802 views; in the same period, a Hyperautomation Labs compilation video got 963 views; matching 11-07 +24 and 11-08 +66.
- Verified English newsletters and directories auto-index it: Enterprise DNA, Substack (Meng Li), The Daily Commit, Reveneau, LinkedIn digest; SkillsLLM, skills.sh, ClaudePluginHub, skillsdirectory, etc.; agents.nix and Yana-AI bundle it
- Inferred The banner is the main source of the long tail's daily gain of 10 stars: code frozen after 09-11, no new posts, yet stars keep coming in at a steady 8 to 15 per day, roughly 2.5% of the big account's weekly star gain. Traffic referrers aren't visible.
Why viettranx/3dviz-pro-max took off
In short: Viet Tran (viettranx, founder of 200Lab), a Vietnamese developer, built 3D visualization agent skills for Claude Code and Codex. The author is also the top contributor to nextlevelbuilder/ui-ux-pro-max-skill, which has 147k stars. It grew mainly on three things:
- Borrowed name: it directly piggybacks on the "Pro Max" name.
- Borrowed traffic: on launch day, a large repo's README got its banner, and it has stayed up ever since.
- Borrowed communities: the author launched on Vietnamese Facebook, then Chinese GitHub recommendation accounts picked it up and reposted it.
After the repo was created the day before (09-11), no code changed, yet stars kept rising steadily by about 10 per day. That long tail is mostly the banner's doing. There's a small spike from 11-07 to 11-08 (+24, +66), matching a Chinese X post and TikTok/Shorts reposts tracked by Github Signals. The product's pitch is that it makes the agent screenshot and look at what it renders, plus a library of 223 recipes. Reposters mostly cite this, but it can't alone carry a 1,100-star repo. As for whether it's the best 3D skill, star counts can't tell us: over the same period, similar projects anatomy (26k★) and img2threejs (21k★) grew far faster.
1. Basic data
- Created 2026-09-10 01:42 UTC, MIT license, JavaScript. All 9 commits fall between 09-10 and 09-11, and the last push was 09-11 01:01. Only 1 external PRs have been merged (#1, J2TeamNNL). There are 2 other issues: #2 is test feedback from the AutoClaw team (09-14), and #3 is an invitation to list it on awesome-ai-plugins (10-31). Data from the GitHub API, checked 2026-11-09.
- Stars: the GitHub API currently shows 1,146★, 129 forks, 9 watchers. Our github library records 1,134★ at 11-09, with gain_1d +72, gain_7d +144, and prev_7d +89. It never made trending. Trending only includes repos at least about 60 days old, and this one was not yet 60 days old, so not charting tells us nothing.
- skills.sh has logged only 176 installs (
npx skills add), a star-to-install ratio of about 6.5:1. This counts only installs through skills.sh, excluding manual copies and plugin marketplace installs (skills.sh, 2026-11-09). - Daily curve:
- Trendshift daily star gains: 09-10 +15,09-11, +40,09-12, +32,09-13, +11,09-14, +45, 09-15 +139 (that day it reached #12 on Trendshift's JavaScript chart), 09-16 +26,09-17, +25. From then until 11-06 it gained +4 to +30 per day, mostly around 10. 11-07 +24, 11-08 +66.
- Over the first 61 days Trendshift shows only +1,028 cumulative, about 115 less than actual. Reveneau captured 09-25★ on 545, but Trendshift had only 09-25 cumulative by 431, so the first-day real figure was roughly 130★; stars gained before Trendshift's first crawl aren't counted.
- Our github table (repo_day): 10-09 693 → 10-28 948 → 11-03 1,002 → 11-08 1,062 → 11-09 1,134.
- Stargazer details weren't retrieved: the endpoint returned 404/401, so individual starring accounts can't be inspected. The fork-to-star ratio is about 11%, the same as the median for new repos of similar size (about 11%), with no sign of star-buying.
2. Competitors
The "coding agent does 3D" niche is crowded, and since 2026 several repos with over 10k stars have appeared from 7 through October. 3dviz is only a small-to-mid player in that field.
| Name | Where | Launched | Scale (github table / API, 11-09) | How it differs |
|---|---|---|---|---|
| wheresryan22/anatomy | Open source | 2026-10-08 | 25,985★, +2,268 over 7 days | Focuses on isometric cutaway illustrations (SVG plus optional WebGL): narrow subject, strong visual impact; 3dviz covers general 3D scenes |
| img2threejs/img2threejs | Open source | 2026-07-15 | 20,924★ | Rebuilds procedural Three.js models from reference images; input is an image. 3dviz takes a one-line prompt. |
| ahujasid/mcp-for-blender | Open source | 2025-03-07 | 33,652★ | Drives Blender directly over MCP: a tool interface with no aesthetics or knowledge base |
| oso95/scroll-world | Open source | 2026-07-06 | 10,883★ | Turns brands into scrollable 3D landing pages; positioned for marketing pages |
| CloudAI-X/threejs-skills | Open source | 2026-01-19 | 4,076★ | 10 Three.js API cheat sheets: covers the API, not aesthetics, with no screenshot verification |
| majidmanzarpour/threejs-game-skills | Open source | 2026-06-14 | 2,755★ | Focused on Three.js mini-games |
| elvarofficial/teardown | Open source | 2026-10-30 | 2,640★, +1,784 over 7 days | A 3D simulation where you take the machine apart, one specific mechanic, climbed past 3 downloads in 10 days |
| arjun988/blender-skills | Open source | 2026-07-05 | 2,234★ | 94 dedicated Blender skills, running through an MCP production pipeline |
| achimala/dream-loop | Open source | 2026-09-07 | 1,846★ | Launched 3 days before 3dviz; generates visuals using Blender, image generation and a subagent review loop, also emphasizing "looking at the picture" |
| Moh4696/open-3d-skills | Open source | 2026-10-31 | 306★, +201 over 7 days | 8 3D web effect skills, also checked by screen recording; positioned for web effects |
| Anthropic's official skills / Claude Code's built-in frontend capabilities | Big-company features | From ¥2025 | anthropics/skills, about 188k★ | Official general-purpose frontend design, with no dedicated 3D knowledge base |
Verdict: It's neither the first nor the biggest. Its position is "broadest subject matter": 24 directions, from fantasy villages to anatomy, linear algebra and robots. It also ships a "honest report" principle, separating what was done from what was actually seen. The fastest risers in the category are the narrow-subject, instantly readable ones (anatomy, teardown).
3. Growth tactics (by time)
- Author's Facebook launch (09-10). Viet Tran's Facebook page has 21,438 followers. 09-10 he posted "Share skill mới: 3D Visualize Pro Max", with the line "Xưa có UI UX Pro Max, nay có 3D Visualize Pro Max cùng mindset" (roughly: "There used to be UI UX Pro Max, now there's 3D Visualize Pro Max with the same mindset"), link in the comments. The post got 253 shares and 24 comments (Facebook post page, via Jina). Follow-on activity in the Vietnamese circle: J2TeamNNL submitted PR #1 the same day; the Vietnamese YouTube Shorts channel "Mint Moment" posted a video with 09-11 views 228.
- Banner in a large repo's README (since 09-10, still up). 09-10 13:34 (+0700): mrgoonie (Duy Nguyen, creator of ClaudeKit and GoClaw, 33k Facebook followers), another core maintainer of
ui-ux-pro-max-skill, pushed 4 times to add a "Check Out Our New Skill: 3Dviz Pro Max" block at the top of the README, with a wordmark, animated GIF and website link (git log -S3dviz -- README.md: 53b8e69, e7f7850, 9cf21ae, 7f69fed). That repo has 147k stars and keeps gaining about 3,000 per week (github table gain_7d 3,092), and the banner is still up today. NextLevelBuilder's website also gave it a product page (nextlevelbuilder.io/products/3dviz-pro-max), and the official X account @nlb_io posted the same copy.- Inference (unverified): 3dviz hasn't changed code in two months, yet it steadily gains 8 to 15 stars per day, roughly 2.5% of the big repo's weekly gain. The most plausible explanation is referral traffic from this permanent banner. The repo's Traffic page requires push access, so this can't be confirmed directly.
- Chinese GitHub recommendation accounts relay (09-11 to 11-07).
- @QingQ77 (Geek Lite) posted 09-11 12:25 UTC, 1,935 views.
- @GitHub_Daily posted 09-14 10:00 UTC, 10.3K views, 20 reposts, 122 likes, 160 bookmarks. The copy reads: "What makes it different is that it forces the agent to look at what it renders". Right after, 09-14 +45 and 09-15 +139 marked the peak of the whole curve; that day it reached #12 on Trendshift's JS chart.
- Followed by @Jolyne_AI (09-18), @hank_aibtc (09-29,193 views), @Smartpigai (10-24), @denziideng (11-07 00:25 UTC).
- English aggregators, newsletters and reposting accounts (passive pickup).
- Enterprise DNA AI Pulse (09-14), Meng Li's Substack "Give the Agent a Skill: Generate 3D Scenes!" (09-15), The Daily Commit (09-16), Alessandro Del Rosso's LinkedIn "GitHub digest" (09-16), French X account @clxymox (09-16), Reveneau (09-25).
- Several small YouTube channels with tens to hundreds of views: LookOnThisGit, FUBAR AI (title "Free 3D agent skill replaces $12/mo Spline"), TheSkillDrop (323), Russian channel Alex To Go, Spanish channel BarbuPrime. A Russian Shorts creator, "Николай Алексеев", drove engagement with "write 3D in the comments and I'll send the link", reaching 1,125 views.
- TikTok repost account Github Signals (11-07). The account has 26,198 followers. 11-07 at 20:14 UTC it posted a video for "3Dviz Pro Max": 3,802 views, 149 likes, 169 saves on TikTok. The same clip got 116 views on YouTube Shorts, and it also went up on Instagram Reels. 11-08 Hyperautomation Labs then posted a compilation video, "I Tested the 12 Free Claude Skills…", which got 963 views. Add those two posts and Denzii's X post, and they line up with +11-07 24 and +11-08 66 on Trendshift, plus +11-09 72 in our database snapshot. The numbers are small, so this is only a minor spike.
- Auto-indexed by directory sites. SkillsLLM, skills.sh, ClaudePluginHub, awesomeclaudeplugins, skillsdirectory (Grade A), claudemap, flowy, gameskill.dev, grokstacks, viralagent.si, MagicNetWorld, Trendshift, and more. agents.nix and Yana-AI even bundle entire skills unchanged into their own distributions (Yana-AI PR #342,805 files).
- SkillsLLM sells featured slots at $49/$99, but there's no evidence the author bought one.
- Missed the big channels: HN (0 entries in our hackernews library), Product Hunt (0 entries in the producthunt library), Reddit threads. No sign of paid ads either.
4. Why it took off
- Borrowed brand and traffic: the name "Pro Max" reuses a hit with 147k stars. The author's Facebook post opens with "we used to have UI UX Pro Max". The big repo's README banner went up on launch day and never came down. That gave it about 130★ on day one and a steady 10 stars a day for two months. It kept growing on days with no code updates and no new posts.
- The author has a community: 200Lab is a programming bootcamp brand in Ho Chi Minh City. The author's Facebook has 21k followers, and the launch post got 253 shares. The Vietnamese developer community was the first ring of spread.
- Chinese recommendation accounts amplified it: a GitHubDaily post with 10k views drove the biggest single day (+139). Almost all the Chinese posts copied the README's numbers (223 recipes, 440 knowledge items) and added a hook: "make the agent look at its own rendered output".
- Heavy packaging: the README has a stats table (223 recipes, 440 knowledge items, 22 kits, 37 studies, 3,599 evidence records), plus a standalone site 3dviz.dev and 37 demo pages you can open directly. Those numbers work well for reposting, and aggregator sites copied them almost verbatim.
- Hits a real pain point: agents writing Three.js often produce a black screen or a pile of boxes, and still say "done". The skill ships a Playwright screenshot script and makes the agent revise against real frames (SKILL.md, README). Substack, GitHubDaily, Denzii, and Github Signals all treated this as the main selling point.
5. Differentiation, and whether that's why it broke out
- Differentiation:
- Broadest topic coverage: 24 opportunity areas, covering creative scenes as well as teaching and scientific visualization.
- Screenshot acceptance: render real frames and revise, instead of trusting the agent's word.
- Honest reporting: separates "what was done" from "what was actually seen", requires sourced facts, and stops animated decoration from passing as simulation.
- Checkable data: recipes, the knowledge base, and the evidence archive recompute their numbers automatically at build time.
- But "look at your own output" is not unique to it. dream-loop (3 days earlier, with subagent review), open-3d-skills (screen-recording checks), and anatomy all do something similar. And the most eye-catching animation in the README's first screen is marked by the author as "an old project recorded before the skill existed, not output from this skill", which weakens the first screen's persuasiveness.
- Verdict: the differentiation gave reposters something to say (every repost cited the screenshot check). The star count, though, came from distribution (the big-account banner, the author's Facebook, Chinese recommendation accounts). In the same period, anatomy and teardown, with narrower topics and visuals that hit instantly, grew much faster. In the 3D skill space, a "broad, comprehensive knowledge base" was not in itself a growth engine.
6. How many people have seen it
- Spreading in small circles. Reach came mainly from Vietnamese Facebook developer groups, Chinese GitHub recommendation accounts on X (GitHubDaily the largest, 10k views), a batch of English newsletters and directory sites, and TikTok repost accounts (3,800 plays). No press coverage, and no front-page placement on HN, PH, or Reddit. It is open source and free, so there is no revenue to speak of.
- The big-account README banner means many developers still pass through every week, but the star count shows very low conversion (roughly 70 to 80 stars a week).
7. Copycats
- No one copying its approach has turned up.
xuenuo126/3dviz-pro-max(10-26 created, 0★) is a single-file 3D visualization of Chinese fortune-telling that only borrows the name.hstptcn5/3dviz-pro-maxis a plain fork. - The "XXX-pro-max" naming has been borrowed by many: vibecode-pro-max-kit (1,167★), editor-pro-max, gpt-image-2-pro-max, and others. But these copy ui-ux-pro-max, not 3dviz.
8. What others can borrow
- Give new launches the banner slot of a hit: treat the top of a 147k-star repo's README as a permanent ad slot, so a new repo gets steady traffic every day even without updates. You need a big account of your own, or an alliance with a big account's owner (here, the same team).
- Inherit the name: "the second Pro Max from the same mindset" lets existing users recognize it as part of a series at a glance.
- Write the README as a repost card: a stats table of "how many X we have", plus a pain-point hook ("the agent says it's done, but the screen is black"). Chinese recommendation accounts and repost accounts copy it verbatim.
- Launch sequence: the author's Facebook, the team's X account, and the big repo's README all went live the same day. Chinese recommendation accounts followed three or four days later, and TikTok repost accounts added a wave about two months on.
- Lesson from the other side: a broad, comprehensive knowledge base can't sustain a surge, and if the first-screen material isn't output from the product itself, it needs a disclaimer, which weakens the spread. The hits in the same period each picked one narrow, visually striking concrete use.
9. Uncertain
- How much traffic the banner actually brought: the traffic referrer data needs push permissions, so it can't be verified. The stargazer breakdown isn't available either.
- The like count on the Facebook post, and whether mrgoonie reshared it to his Facebook (33k followers), could not be checked.
- How the 11-08 66 stars split among Denzii, Github Signals, and Hyperautomation Labs can't be separated out.
- Whether the author paid for a featured slot on SkillsLLM is unknown.
- Why anatomy and teardown grew faster was not investigated this time.
Sources
- GitHub API: repos/viettranx/3dviz-pro-max (commits, issues, forks), users/viettranx, repos/nextlevelbuilder/ui-ux-pro-max-skill (contributors, README;
git log -S3dviz) - Our github repo_day and repo_profile tables (10-09 to 11-09); hackernews and producthunt libraries (no records)
- Trendshift trendshift.io/repositories/226315
- skills.sh www.skills.sh/viettranx/3dviz-pro-max
- Viet Tran's Facebook post www.facebook.com/viettranx/posts/29345405191727554/
- X: x.com/GitHub_Daily/status/2099438031858270430, x.com/QingQ77/status/2098387330314699172, x.com/denziideng/status/2118862320504082890, x.com/hank_aibtc/status/2104865865506357718, x.com/nlb_io
- TikTok: Github Signals, video 7705525418605939986 (TikHub); YouTube Data API (vRvdDhOD1C8, ahkfL1qZleE, fh42bmHrdGk, etc.)
- pythonlibraries.substack.com/p/give-the-agent-a-skill-genera…, reveneau.com/ainews/viettranx-3dviz-pro-max-claude-code-code…, thedailycommit.in/story/2026-09-16/08-github-viettranx-3dviz…, nextlevelbuilder.io/products/3dviz-pro-max, 3dviz.dev/